Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
New Credit: How Data Science is Changing Lending
By: James Jasper and Jared Maslin
Final Draft: 5/04/2015
1
Introduction
Since the financial crisis, low-income growth and persistent poor credit has limited many
from achieving stability in their lives. These dual limitations has led to their being 50 million
individuals deemed “unscorable” by traditional credit methods – effectively freezing them out of
the market for home loans, car loans, and access to credit cards.1 Alongside these developments,
technical hurdles to accessing social networks and other methods of rating an individual’s ability
to repay a loan have decreased. Data science is now being used to extend a variety of alternative
credit instruments to these underserved individuals.
While we applaud the effort to reach this population with needed financial services and
new options for credit and debt consolidation, using data science’s methodologies in lending is
an area ripe for abuse, likely illegal due to violations of the Equal Credit Opportunity Act and the
Fair Credit Reporting Act, and discriminatory. In this paper we will examine how the current
state of credit markets came to exist, how data science is now involved in lending decisions, and
the associated regulatory and ethical considerations of its use. We find the current justifications
of consumer credit practices by non-insured institutions to be unsettling due to the lack of
transparency in their credit decisions and the consistently higher than market rates they charge to
borrowers. At the end of this paper we suggest improvements to current legislation that can
protect consumers and companies alike.
History of Credit
Credit is the foundation of modern society. The consistent ability to rate a borrowers ability to
repay a loan reduces risk in the overall economy and provides lenders a relatively safe store for
1 Crosman, Penny and Andy Peters. “New Underbanked FICO Score Faces Old Banker
Skepticism”. April 2, 2015. http://www.americanbanker.com/news/bank-technology/new-
underbanked-fico-score-faces-old-banker-skepticism-1073600-1.html.
2
excess savings. More importantly, access to credit is the foundation of a fair and equal society. A
critical component of lending decisions, the concept of credit has existed since the ancient
Greeks.2 The extension of credit is based on the ability of an individual to repay – originally a
decision was based on the relations of that individual, amount they desired to borrow, and
income or assets they had against the loan. A rudimentary credit score was used to define the
terms of any deal. As such, credit and credit scoring evolved and became more sophisticated to
accommodate changes in the banking and to address the needs of society. Many of the laws that
advanced the rights of minorities centered on this access; the Fair Housing Act (FHA) of 19683
and the Community Reinvestment Act (CRA) of 19774 both set strict requirements on the ways
credit was extended to protected classes and the obligations of banks to lend into the
communities where they were based.
However, since the passage of both the FHA and CRA technology has rapidly outstripped
the ability of the government to regulate the banking industry. The advent of the FICO score in
1989 by Fair, Isaac, and Company standardized the assessment of the ability to repay. The use of
the FICO score greatly reduced the abuses of the credit system as it removed the ability of banks
to use arbitrary interviews to deny credit. Additionally, the proliferation of the Internet
broadened the markets available to banks. As such, a local bank could now operate across state
lines for clients while providing the same level of service. Traditional methods of client relation
building broke down as ease of access to information increased. Large banks became
increasingly centralized, as they would lend to any qualified borrower.
2 Homer, Sidney. History of Interest Rates: 2000 B.C. to the Present. New Jersey: Rutgers U.P.,
1963. 3 42 U.S. Code Chapter 45, Subchapter I. https://www.law.cornell.edu/uscode/text/42/chapter-
45/subchapter-I.
4 12 U.S. Code Paragraph 2901. https://www.law.cornell.edu/uscode/text/12/2901.
3
This growth and acceleration of the banking sector carried certain risks that were not
fully understood until 2007 when the housing market collapsed. While credit had built a great
number of homes, it just as swiftly resulted in a collapse and freeze of credit markets from which
we have yet to fully recover.5 For consumers, the results of this collapse have been two-fold:
1. The portion of the population with subprime or near prime credit rose to 56% of the
population.6
2. The portion of unbanked individuals stands at 7.7% and underbanked individuals at 20%.
To put that in perspective, nearly half of credit users would have difficulties getting a home or
auto loan. Nearly one-in-four do not have access to credit cards, checking or savings accounts,
and as a result are stuck in high cost alternative financial services.7
The combined effects of consistently depressed credit scores and a mistrust of traditional
financial institutions have resulted in a fundamental shift in how lending services are being
provided to lower income individuals. Alternative Financial Services (AFS), services provided
5 As defined by failure to achieve stable job creation and income growth which are defined by
the Federal Reserve Board as the key elements influencing their decision to keep interest rates at
historic lows. 6 Defined as consumers with a 700 or lower FICO score. For a full list of data: “Financial Assets
and Income: Consumers with Subprime Credit.” 2015 CFED Survey.
http://scorecard.assetsandopportunity.org/latest/measure/consumers-with-subprime-credit. For a
fuller explanation see: 2013 FDIC National Survey of Unbanked and Underbanked Households.
October 2014. https://www.fdic.gov/householdsurvey/2013report.pdf 7 “Alternative Financial Services: A Primer”. FDIC Quarterly 2009, Vol 3:1, 39-47.
https://www.fdic.gov/bank/analytical/quarterly/2009_vol3_1/FDIC140_QuarterlyVol3No1_AFS
_FINAL.pdf
4
outside of federally insured banks and thrift institutions, now account for $320 billion dollars of
transactions annually8:
Furthermore, traditional institutions have little interest in attempting to acquire unbanked and
under banked individuals. There is a woeful lack of educational material, understanding of
financial services, and access to checking and savings accounts to this traditionally underserved
population.9
As a result of these shifts, there has been a significant focus by private equity and venture
funded firms to provide services to this underserved population. However, with extension of
8 For a complete assessment of the impact of Alternative Financial Services see the FDIC’s
primer on AFS above. 9“FDIC Survey of Banks’ Efforts to Serve the Unbanked and Underbanked: Executive Summary
of Findings and Recommendations.” Feb. 2009.
https://www.fdic.gov/unbankedsurveys/unbankedstudy/FDICBankSurvey_ExecSummary.pdf
5
credit through AFS style investment vehicles are high interest that further disadvantage
borrowers who already have low incomes and difficulty repaying.10 Data science has been on the
forefront of these lending decisions – using payment history and open lines of credit to determine
a borrower’s likelihood to repay. However, more recent advances in the field of data science
have allowed lenders to reach much deeper into an applicant’s histories to determine ones
creditworthiness.
Credit and Lending
Statistics and probability has always been used in lending to some extent. The market for debt is
determined by future expected returns and the related risks to a given individual or company.
Only recently, however, has statistics and probabilities been used at the individual level through
the advent of a standardized FICO score from the Fair Isaac Corporation in the 1980’s.
Since the advent of a numeric score for credit reporting, the practice has become
widespread with a variety of other companies offering similar scores.11 The premise behind these
scores is the use of predictive modeling to identify the risks of default of the borrower on a loan.
As such, different lenders can establish different criteria in terms of trade lines and payments on
those lines for loans based on the risks they are willing to take and price their products
effectively for that risk.12 There are positive and negative aspects to using a unified approach to
risk scoring. Positive results include:
10 Klepper, David. “NY lawmakers examine subprime auto loans.” April 23, 2015.
www.washingtontimes.com/news/2015/apr/23/ny-lawmakers-examine-subprime-auto-loans/ and
Harnett, Sam. “One Uber Driver’s Story: How He Was Trapped by Auto-Loan Program.” April
24, 2015. ww2.kqed.org/news/2015/04/24/one-uber-drivers-story-how-he-was-trapped-by-auto-
loan-program. 11 “The Impact of Differences between consumer- and creditor-purchased credit scores.” CFPB.
July 19, 2011. http://files.consumerfinance.gov/f/2011/07/Report_20110719_CreditScores.pdf 12 It should be noted that these standards are often opaque themselves: many large banks require
multiple trade lines open for multiple years. These requirements aren’t typically disclosed
6
Ease of comparison between borrowers.
Effective risk based pricing for individuals.
Simplified transaction cost through ease of access to credit history across lenders.
High degree of regulation with Fair Credit Reporting Act Reg. B and Reg. V.
Negative outcomes include13:
Heavily dependent on length of credit history (35% of score).
o Bias against younger people and people new to credit instruments.
Does not account for assets or income, only payments.
Focus on percent utilization of available credit as opposed to ability to afford spending
levels.
Minority neighborhoods average 550 credit scores, non-minority neighborhoods average
a 700 score.
As previously noted, a poor credit score drastically increases the cost of available financing and
can freeze someone out of traditional financial institutions forcing them into costly alternative
financial services.
New Risk Calculations Using Data Science
The combination of the effects of a large population of subprime borrowers and a large
group of unbanked individuals has allowed more advanced methods of data science to make
inroads into lending. This significant opportunity has attracted many data science and analytics
upfront to a prospective borrows, but would be disclosed in an adverse action notice as described
by Regulation B of the Fair Credit Reporting Act: 12 CFR 202—Equal Credit Opportunity
(Regulation B). Office of the Comptroller of the Currency. U.S. Department of Treasury. July
15, 2011. http://www.occ.gov/news-issuances/bulletins/2011/bulletin-2011-39.html. 13 Meeks, John. “FICO Scores”. FDIC. October 28th, 2011.
http://www.wvasf.org/presentation_pdfs/John_Meeks_-
_WV_Asset_Building_Charleston_102811.pdf
7
companies into consumer lending. These companies vary in style and organization but their
overall strategies are clear: increase the number and the variety of parameters in a lending
decision to reduce the risks.
As far as lending goes, the number of companies has rapidly multiplied: AvantCredit,
Moven, Kabbage Inc., Kreditch, LendUp, Finca, Neo, ZestFinance, LendingClub, and Lenddo
are just a few companies adopting this approach. While some focus on consumer lending such as
ZestFinance and LendingClub, others like Kabbage, lend to small businesses. Data gathering
focuses range from examining LinkedIn and Facebook profiles for number of connections to
gaining direct access to bank accounts and spending habits.
To a certain extent, such activities are a natural outgrowth of the growth of available data
online. In a business context, Yelp reviews as well as web presence are important factors in the
success of a modern business. However, it becomes much more tenuous to suggest that lending
decisions can or should be based on online profiles. Recent articles have noted the broad fraud
that social media has allowed; fake profiles abound, look legitimate, and are difficult to stop.14
As such, there’s a large possibility of fraud that could be perpetrated by borrowers to dupe
lenders. Additionally, legal and ethical considerations for the lenders are a serious concern as
they should be able to justify lending decisions that they are making for individuals.
Legal Concerns
At the core of legal considerations around the extension of credit is the Fair Credit
Reporting Act (FCRA) and the Equal Credit Opportunity Act (ECOA). The FCRA stipulates that
credit reporting to consumer credit agencies must be accurate and the note of adverse credit
14 Bilton, Nick. “Social Media Bots Offer Phony Friends and Real Profits.” Nov. 19th, 2014.
http://www.nytimes.com/2014/11/20/fashion/social-media-bots-offer-phony-friends-and-real-
profit.html?_r=0.
8
decisions, in the form of a denial of financing, must be provided with reasoning. Additionally,
the ECOA provides that there should be no discrimination in access to financing based on age,
race, religion, or any other arbitrary factor. The FCRA provides a broad definition of consumer
report agency or consumer report as well:
FCRA, 15 U.S.C. § 1681a(d): any written, oral, or other communication of any
information by a consumer reporting agency on a consumer's credit worthiness,
credit standing credit capacity, character, general reputation, personal
characteristics, or mode of living which is used or expected to be used or
collected in whole or in part for the purpose of serving as a factor in establishing
the consumer's eligibility for (A) credit or insurance to be used primarily for
personal, family, or household purposes; (B) employment purposes; or (C) any
other purpose authorized under section 604.
Information in such reports should be accurate and redress exists for consumers who
notice inaccurate credit information on such reports. The number of organizations that could fall
under this FCRA umbrella is quite large.
Based on the known methods of how these new firms extend credit, some major legal
concerns come to the fore. First and foremost is the adherence to the FCRA and ECOA in whole.
While this may seem to be simple, the Regulation B of FCRA significantly complicates the
extension of credit in this way. Regulation B states that after an adverse decision is made proper
notice of why credit was not provided should be provided once receipt of a full application has
been achieved. Given that many of these algorithms examine large numbers of data points, such
a provision would be difficult to provide – stating that the user did not have enough Facebook
friends while an equivalent user may have qualified with the same number of friends would not
9
be an understandable difference. In fact, such a decision would be considered arbitrary as one
would need to prove that this factor significantly increased risks to the lender in a measurable
way.
To further complicate matters, if social networks are being examined, this data could
violate the ECOA. For example, research has shown that social networks are vitally important in
determining future earnings. If a company is now deciding to provide credit based on social
networks, there has shown to be significantly more earning potential among white individuals
due to their networks than African Americans.15 Additionally, social networks self-segregate
along racial lines.16 One could violate the ECOA by examining a social network, and while race
may not be openly available, associated information would allow the inference of this
information. Rather than making a credit decision on concrete facts, the results would be based
on inferential information – another violation of current legal statures. Such data mining would
be in violation of the ECOA as it clearly defines a lending decision along racial lines as illegal.
However, these concerns must be balanced against the benefits that these services
provide to unbanked and underbanked individuals. For instance, if no one can achieve financing
through traditional methods, the legal considerations that the FCRA and ECOA are supposed to
address have failed to succeed. The question then becomes to what extent should innovation in
finance be allowed versus how much current regulations need to be changed to serve other
individuals.
15 Flores-Yeffal, Nadia Y. and Li Zhang. “The Role of Social Networks in Determining
Earnings: A Comparison Analysis of Four Racial and Ethnic Groups.” Sociology Mind: Vol 2, 5,
2012. 235-246. http://www.scirp.org/journal/PaperDownload.aspx?paperID=18447 16 American Values Survey. October 29th, 2013. http://publicreligion.org/research/2013/10/2013-
american-values-survey/#.VT2jNK1Viko
10
In addition to the legal considerations, much of this data analysis crosses the lines into
research on human subjects. As such, there are ethical concerns in additional to legal
requirements. We try to apply the Belmont Report to these concerns.
Applying the Belmont Report
The Belmont Report, developed by the National Commission for the Protection of
Human Subjects of Biomedical and Behavioral Research, puts forth a framework of ethical
guidelines that are meant to aid in ensuring the protection of human subjects in research. Despite
being developed in 1979, the report has been upheld over time as an essential rubric in all areas
of research. Data science, in many ways, employs many of the same experimentation and
procedural methods as more traditional areas of research, such as physiological or medicinal
research. When medical research might involve testing the effects of a specific drug or treatment
onto a set of subjects, data science research can also involve the use of personal data from human
subjects, whether in a financial, psychological, or other realm. With this in mind, this paper will
consider the potential applicability of some key aspects from the Belmont Report onto the
current framework of social data usage in the lending process.17
The Belmont Report details three primary areas of ethical focus concerning what is
necessary to ensure an ethical research practice including respect for persons, beneficence, and
justice.
Respect for Persons
The principle of “respect for persons” focuses on the relative capacity or vulnerability of
each individual subject being included in a research project. Tantamount to this is the capacity to
17 National Commission for the Protection of Human Subjects of Biomedical and Behavioral
Research; “Ethical Principles and Guidelines for the Protection of Human Subjects of Research”;
http://www.hhs.gov/ohrp/humansubjects/guidance/belmont.html., April 18, 1979.
11
make decisions or contemplate issues that could arise both before and during research. In
pharmaceutical research, this could be viewed as the capacity of a trial participant in considering
the risks and benefits of their involvement in a research project and, if necessary, contemplation
of removal from the project should they deem their participation to be inappropriate or harmful
in any way.
In applying this principle to the realm of lending, one might view respect as a necessity
to assess each participant’s potential ignorance to an institution’s use of this data and the impact
that it can have on their lending decision. While some subjects may be fully aware of the
potential consequences of the use of their social data in lending decisions, others may not fully
comprehend the impact that it can have, depending upon what the lender finds and how they
choose to interpret the information.
Take AvantCredit, for example – Avant offers consumer lending to underserved
borrowers through the use of publically-available information. Avant uses consumer credit
information, which is no different than the vast majority of consumer lenders in the market
today. Where Avant diverges in practice is their use of social data and user action data, tracking
their activity online (i.e., social media websites) and the network of friends, colleagues, peers
that they have constructed through the use of social media applications. While being “personal
data” in the minds of many social media users, the data itself is viewed as publically-available to
many entities, as almost anyone can go online and retrieve the information that Avant is using in
its lending decisions.18
18 Rakesh, Chitra. “Online lender AvantCredit secures $34M in funding”. Venture Beat.
http://venturebeat.com/2013/05/09/online-lender-avantcredit-secures-34m-in-funding/. May 9,
2013.
12
In applying the Belmont standard of respect for persons, one might wonder how Avant is
ensuring that potential borrowers are fully aware and capable of contemplating the use of their
social data in their consumer lending determinations. Are borrowers made aware prior to being
evaluated for a loan or is this something in the fine print shown after the fact? If such a gap were
to exist, companies like Avant would need to address the clarity with which it communicates the
benefits and risks to borrowers of using social data in their decision making.
Beneficence
The principle of beneficence is a fairly simple one, focusing on the reduction of harm to
research subjects and the maximizing of benefits to subjects, to the fullest extent possible. In
many ways, beneficence runs alongside a doctor’s Hippocratic Oath of “do no harm”. The
additional aspect of maximizing the benefit can be thought of as the researcher’s responsibility to
consider all potential benefits of a practice, regardless of the party or parties that could become
beneficiaries.
Taking Avant as an example again, there should be an expectation that Avant considers
and documents the potential risks and benefits to use of social data in lending determinations.
Being a for-profit business, there could be a tendency for management to observe the ways in
which using borrower-related data could increase their overall profitability and nothing further.
Doing so would not satisfy the principle of beneficence, though, as one must also consider the
benefits and risks of the practice for all parties that could potentially be impacted. Could this
practice result in deceptive lending practices or in egregious lending products being offered to
consumers? Could this practice increase economic risk in ways that subprime lending has been
shown to impact economies in the recent history? In these ways, Avant should be responsible for
a full consideration of the entire impact set surrounding their lending practices.
13
Justice
The principle of justice can be seen as a detailed principle with several facets to consider.
At a high level, the principle can be viewed as a set of considerations in spreading the respective
benefits and burdens across the parties involved. The set of considerations include the use of
equitable distribution, need-based consideration, merit-based consideration, levels of relative
effort, and levels of relative societal contribution. In many ways, each of these considerations
points to the relative fairness of a given research or business practice. Ensuring justice in a
lending practice would require that each institution exercise discipline in determining how to
execute their business plan and the impacts of its practice onto its subjects and society, as a
whole.
For an example, let us consider Lending Club, which originated as a social application
within Facebook’s platform. Lending Club uses a peer-to-peer lending process, acting as an
intermediary between potential public lenders and borrowers. In doing so, the principle of justice
requires that Lending Club exercise control in selecting its consumer base according to their
respective needs and merit-based attributes. For this, they could observe the FICO credit score
and income of potential borrowers, while assessing the availability of competing institutions’
financial products to potential borrowers in the market. This will allow for Lending Club to
ensure that it is applying an ethical focus to its operations and to its selection of business
partners.19
In addition, Lending Club must ensure an equitable distribution of the benefits between
the business, its customers, and society, as a whole. One might contemplate whether lenders are
making a fair return on their investment, relative to that which Lending Club secures through
19 Lending Club. “How does an online credit marketplace work?”
https://www.lendingclub.com/public/how-peer-lending-works.action. 2015.
14
each transaction. Also, are borrowers being given a fair interest rate for their loan or is Lending
Club gouging into the pockets of vulnerable or desperate borrowers? Even further, how does
Lending Club’s existence alter the lending market for consumers? Do Lending Club’s practices
oversaturate the market or make it more difficult for consumers to obtain lending from other
commercial sources (i.e., does it drive up rates from other institutions)? The principle of justice
would demand that Lending Club consider and implement an equitable distribution of the
business operational detriments and benefits amongst all parties that are potentially impacted.
Limitations of the Belmont Report as a Business Framework
While the Belmont Report provides a very useful framework in considering the ethical
aspects of a research method for human subjects, its application to businesses are somewhat
limited on a few fronts. For one, many research efforts are not implicitly driven by profitability
as a key motive. Conducting research on the quality of learning for students in various regions or
learning environments may not be principally driven by a desire to make money, in one way or
another. For example, enhancing educational experiences for students and teachers alike could be
a driving motivation to conduct research of this nature. In considering for-profit businesses,
though, profitability is often a guiding principle that leads all business-related discussions of
products, processes, or people.
Matters of social responsibility are debated heavily, with profitability going head-to-head
with a perceived obligation that businesses serve the community first and foremost. The Belmont
Report details matters of accountability and responsibility, but not necessarily those of
profitability. In the lending market today, lenders may dispute that the act of lending places the
majority of risk onto the business’s shoulders, justifying the distribution of profits or benefits to
the lender before the borrower. Even further, the use of social data can be defended as a manner
15
of protecting the business from deception on the part of potential borrowers. One might argue
that through the use of social data, lenders can uncover details about potential borrowers that
they would otherwise wish to conceal when securing a loan. These matters can often transcend
the confines of the Belmont Report and require additional consideration in order to understand
the risks and benefits of a business practice.
Second, many research practices observe risks to subjects as those that can manifest
themselves during the course of a project. For example, if a researcher wants to understand how
multi-tasking affects productivity and seeks to construct an experimental design to attack this
problem, the researcher may consider potential harms like the physical fatigue or
mental/psychological strain that subjects might experience. It might be considered less relevant
to look further down the road to how subjects might be affected one, two, or twenty years down
the road, and in many ways, there isn’t necessarily a reason to do so. With business practices like
consumer lending, the introduction of potentially subjective social data in lending determinations
could have drastic, long-term impacts on how lending is structured in the future and how that
structure impacts the financial market. As seen in recent years, subprime lending practices can
cripple an economy if left unchecked or unregulated. While seemingly harmless, the Belmont
Report lacks an infrastructure to define long-term impacts that can loosely stem from the use of
social data in lending decisions.
A Fine Line – Social Data and Lending Decisions
The lending industry, as it exists today, is intensely regulated and monitored by the
Federal Government. Fair Lending laws like the Equal Opportunity Act 20 and the Truth in
2015 U.S.C. 1691 et seq., 12 C.F.R. pts. 202 and 1002 and 12 C.F.R. 701.31 (NCUA)
16
Lending Act21 seek to protect potential borrowers through ensuring equality and clarity in the
offering and consideration of lending opportunities. The process for seeking out and engaging
potential borrowers is restricted, both in aspects of timing and method. The use of social data,
though, throws a bit of a wrench into the current legal framework.
By nature, social data regularly includes personal details such as race, physical build
(through images), marital status, sexual orientation, religious beliefs, and personal hobbies. And
these are just a few of the possible data points that can be obtained simply through the scraping
of a borrower’s social media homepage. While being publically available, the information
crosses the proverbial “line” and into the realm of protected parties. This begs the question of
how social media data can be used, yet controlled in a way that prevents any kind of
discrimination or bias from being introduced. How might this information be used to target
vulnerable groups or exclude those that are deemed “unfit” for lending on the basis of social
media data? And unless the use of such information is recorded and tracked to an incredibly
detailed degree (which likely wouldn’t be cost-effective or even feasible), how can regulators
ensure that the information is being used in a legal and ethical manner?
The Federal Trade Commission (FTC) Act also presents a potentially troublesome legal
hurdle to consider. Section 5 of the FTC Act protects against “unfair or deceptive acts or
practices in or affecting commerce.”22 One potential concern here is that it can be exceedingly
difficult to prove the intent and eventual use of certain pieces of data. When a researcher scrapes
profile information off of a social media site, it’s incredibly likely that included in their data set
21 15 U.S.C. 1601 et seq.; 12 C.F.R. pts. 226 and 1026 22 16 C.F.R. Part 255, FTC “Guides Concerning the Use of Endorsements and Testimonials in
Advertising”.https://www.ftc.gov/sites/default/files/attachments/press-releases/ftc-publishes-
final-guides-governing-endorsements-testimonials/091005revisedendorsementguides.pdf. 2009.
17
is a mix of personal or potentially discriminatory data. Once that data is gathered, how can
businesses be monitored or effectively regulated to ensure that discrimination or unethical bias in
commerce never come into play? And even if certain data points are gathered, but not
specifically used in decision making, can the public ever truly be comfortable that the process is
devoid of bias and that the data isn’t “tainted” in some way?
Regulatory Compliance and Dubious Legality
Being such a highly regulated industry, each individual lending decision comes with a
laundry list of requirements and expectations in terms of what is communicated, how it is
communicated, and to whom it is communicated. The formal reasoning for a judgment must fit
into a narrow lane, as must the manner in which that decision is documented. But with this
comes an innate weakness to the system – regulators have neither the time, nor the resources to
audit or inspect lending processes beyond the specific wording in legislation. As such, lenders
must focus heavily on “to-the-letter” compliance in their daily operations. What this also means,
though, is that there is likely opportunity to identify loopholes or flexibilities within current
legislation in order to heighten profits or create a competitive advantage.
In many ways, this is what gave rise to subprime lending in the first place, with
legislation being fairly weak (if not unaware) in terms of loan packaging and debt swaps. As
long as the explicit regulations were recognized, lenders use any remaining slack to grow the
bottom line. Though a separate issue, the introduction of social data into lending decisions piques
many of the same concerns. If regulations state that a lender cannot discriminate potential
borrowers due to their race, sexual orientation, national origin, and so forth, then it may be
considered enough to refrain from explicitly doing so in an openly malicious manner. However,
that “letter of the law” today doesn’t necessarily prevent a lender from creating a data-based
18
algorithm, being based off of social data that uses personal information and that of their network
of social contacts in order to determine their credit worthiness. Lenders would simply need to
walk the tightrope, so to speak, as many have done for generations in the name of profits and
business continuity.
The above example highlights the ever-present “dubious legality” in the lending industry,
requiring direct regulatory requirements while using any remaining wiggle room in order to gain
a financial edge. Lenders may argue, though, that this is essence of a business. Higher revenues
and lower costs drive higher profits and returns for all parties involved. What right does the
public have to define how businesses operate or how they work to optimize their lending
practices? So, if no laws are being violated, what’s the real problem? Consider the following
role-play between a lender and potential borrower:
Lender: “We’ve reviewed your file, complete with social information and details of your
technological use as of late, and we’ve come to a decision.”
Borrower: “You reviewed what? What does my social status have to do with my lending
status?”
Lender: “Your background and social activity can offer some insights into your
perceived strengths and weaknesses as a potential borrower.”
Borrower: “So, what? My pictures, friends, and family are all fair game? And my phone,
too?”
Lender: “In no way would we ever discriminate based on any gender, religious, or racial
aspects during a lending decision. The practice simply offers new insights into your
history than a credit score alone.”
19
Borrower: “If there’s no discrimination, why is the rate you’re offering to me different
than what it would be from my FICO credit score, by itself?”
Though a seemingly extreme example, there is a significant potential for this very
predicament to present itself if the practice of using social data goes awry. And if we put
ourselves in the shoes of the borrower, what are they to do at that point? If they’re fed up and
decide to walk across the street to a different bank, what’s to say that the next lender won’t be
using the very same practice? Or who is to say the other lender would extend credit on traditional
factors? It’s in the nature of a business to observe competitor practices and to adapt with
practices of their own, thus combatting a competitor’s advantage in the market. What’s to stop it
from happening if the practice is legal and in compliance with current regulations, but still has a
hint of unethical reasoning embedded within its foundation?
One efficient manner in which to use the social data, as previously discussed, is the
generation of algorithms in order to process data and spit out a lending recommendation as an
aggregate of the data consumed. In one way, algorithms can be seen as an unbiased decision-
making process that simply follows finite instructions in order to reach a conclusion or end
result. To rest on this, however, would be irresponsible from a risk perspective. One must also
recognize that algorithms are developed by people in the industry, injecting industry insights and
procedural logic along the way.
So, while an algorithm may be an unbiased processor, what can be said about its
designer(s) and their original logic? In the same manner as a statistical test, if a hypothesis or
experimental design is biased in its construction, then any results from that process can be
flawed, as well. Therefore, any bias or discrimination that leaks into “industry logic” could
eventually flow into the design or implementation of their algorithms. This is yet another
20
example of how a lender, though in compliance with lending regulation, can potentially be
executing in an ethically “gray area”. It is a constant game of tug-of-war between a consumer’s
quest to save money and a business’s quest to increase their yield.
Public vs. Private Capital
A broad departure in the lending risk environment can come from a lender’s decision to
use either public capital or private capital in order to back their financial instruments. The
manner in which these businesses operate, are led, and handle financial decisions can be wildly
different. In some ways, it all depends on the source of financial backing and the regulatory
requirements related to their business’s structure. A large majority of the banks hit the hardest by
the subprime lending fiasco over the last decade operated as public entities. In most cases, they
were public investment firms that were traded on the New York Stock Exchange (i.e., Goldman
Sachs, Wells Fargo, Wachovia, Merrill Lynch, etc.).23 Any lending, selling of packaged loans, or
credit default swaps were regularly backed by the Federal Deposit Insurance Corporation (FDIC)
insurance policy to some degree. Public funds and publically-owned debt were frequently
packaged as marketable financial instruments.24 As such, this spread some of the procedural and
financial risks onto the public, rather than on the business itself. The Federal Government does
what it can to protect the public from poor business practices, but the big banks of the world
simply have to high of volume to follow around with a magnifying glass.
Private capital, on the other hand, presents a more internally risk-based environment, and
as such, can result in lower government scrutiny in their lending practices. For this reason, we
have seen a jump in recent years in the number of private equity lenders entering the market. The
23 Labaton, Stephen. “Agency’s ’04 Rule Let Banks Pile Up New Debt”, The New York Times.
http://www.nytimes.com/2008/10/03/business/03sec.html. October 2, 2008. 24 Federal Deposit Insurance Corporation. https://www.fdic.gov/about/index.html#1. 2015.
21
idea behind private capital presents the potential for lower government involvement and lessened
reliance on the public to define its business practices. This is not to say that private equity firms
can do whatever they want, but instead, reflects a simple distinction in the placement of risk. For
a private equity firm, lending decisions and financial instruments are backed by existing private
capital, rather than governmental or public funds. Lending Club is a fantastic example of this
model, with peer-to-peer lending that is backed by privately-held investments and not federal
deposit insurance. As lending determinations become more and more complex (or creative,
depending upon your view), we will likely continue to see private equity firms enter the market
in hopes of employing more sophisticated operating models to the industry.
A New Model Proposition
Existing regulatory models for financial institutions, while extensive, are ill-equipped to
deal with an incredibly dynamic environment like data-based consumer lending. There are
simply too many facets within the industry for a single regulatory body to monitor all aspects of
the lending process, especially with the advent of social data feeding into lending determinations.
Where legal requirements may be fully obliged, there are many privacy-related and ethical
implications stemming from the latest trends in lending analytics. As seen with many other
industries, it is common for innovation and procedural creativity to outpace legal and ethical
frameworks. Uber and Lyft are perfect examples of private industry outpacing public regulations.
With this thought in mind, an overarching ethical framework should be constructed in a manner
that forces business to add transparency and accountability to their operations.
The following model is somewhat akin to that which is employed for the Sarbanes-Oxley
Act of 2002, which was developed in response to financial reporting and fraud-related scandals
22
in public business like Enron and Worldcom.25 The proposed model below is intended to be used
in both public and private lending environments. Simply put, the model involves a cyclical
course of action that allows for accountability and transparency, as well as opportunities to
examine lending practices and adapt to the needs of a dynamic industry.
The process is intended to be iterative, as doing so allows for the business to adapt while
potentially gaining as much from the process as those that seek heightened control and
transparency in lending practices. Instead of an internal control design, however, this process
would involve the formal documentation of social data usage from all aspects of the business.
This would include the retrieval and storage of data along with any subsequent analytics and
decision-making guidance that stems from the analysis.
The first phase of the process would involve a requirement to fully document the design
of its data and decision making model from womb to tomb. As much as it would involve a visual
flow of information through the process, it should also include a written description of what data
is being gathered, what data is being used, how data is being used, and how data is being secured
with retention plans. One must recognize, though, that many lenders could be skeptical about
documenting their processes too intensely, as it could indicate or hint at the source of any
25 107 Congress Public Law 204 (Sarbanes-Oxley).http://www.gpo.gov/fdsys/pkg/PLAW-
107publ204/html/PLAW-107publ204.htm. 2002.
23
competitive advantage and give other lenders an edge. The nature of this process is that it would
be universally applied to the lending community and that any minute, proprietary details could
be maintained. It is more of a focus on the commitment to fair and honest lending, free of bias or
discriminatory treatment. The design documents created during this phase would be made
available to potential borrowers in considering business with the lender, in order to ensure that
consumers are given full disclosure in route to an informed decision.
While some institutions could argue that their terms of service documents, which are
available online for many leading institutions, detail this information today, there is a significant
difference in the degree of the detail being discussed. Many terms of service are exceedingly
vague, which allows for a lot of underlying flexibility in how that agreement is applied in the
lending process. Zest Finance, for example, is an innovative payday lending organization whose
process is rooted deeply in the analytics of big data. Their website details their motto of “All
Data is Credit Data.”26 Furthermore, their policy indicates the following: “With a team of some
of the world’s best data scientists from Google and lending experts from Capital One,
ZestFinance analyzes thousands of potential credit variables – everything from financial
information to technology usage – to better assess factors like the potential for fraud, the risk of
default, and the viability of a long-term customer relationship.” What is not evident, however, is
to what extent the data gathering process will go and how it will be used.
Again, one recognizes that some of the answers to this question would indicate
proprietary logic. What is being proposed here is for financial institutions to bridge the gap.
Meeting consumers in the middle to explain what data is being used and to what extent, without
26 ZestFinance, “How We Do It”. http://www.zestfinance.com/how-we-do-it.html. 2012.
24
revealing the proprietary calculations in detail, could offer a great deal of transparency into the
existing fairness and justice in the lending process.
The second phase of implementation involves executive leadership as well as internal
functions, ensuring that the process is put into place according to the documents created during
the design phase. Without a directed commitment to transparency and follow-through during
implementation, the design phase could be rendered moot and borrowers could return to the dark,
where lending practices and handling of personal data is largely unknown or ambiguously
reported.
The third phase of monitoring is one that is most beneficial to the lending organization, as
it is an internal function and allows for the identification of procedural opportunities and
improvements. If through the course of normal business, management becomes aware of an
adverse reaction to the treatment of certain social data sets, a formal monitoring task would be
essential in identifying opportunities to improve processes and react to consumer demands as
quickly as possible. In the same vein, if social data is being gathered, but not utilized to its
potential in a lending determination algorithm, a monitoring effort would allow for adjustments
to be made in a clear and efficient manner. Also, an iterative process of this nature ensures that it
can be utilized as a continuous improvement method by cycling through the process in
perpetuity, increasing awareness and accountability to its customers. It eases notions like
informed consent and documents the usage of data in its processes so that before entering into an
agreement, consumers can fully understand (at least to a reasonable extent) the implications of
the process and how a determination was made. Though an extremely delicate matter, the
proposed process would benefit all parties involved with increased transparency, reliability, and
consumer understanding in the lending industry.
25
Conclusion
Data Science has the potential to extend credit to a greater number of people and provide
access to credit markets to many underserved individuals. However, a focus on ethical and legal
concerns regarding these credit decisions needs to occur before such technology can truly go
mainstream. We propose extending existing legal and ethical frameworks through iterative
review processes that incorporates existing legislation but allows room for innovation in lending.
Additionally, we advocate continued application of existing laws to ensure equitable and fair
access to credit markets for all individuals.
26
Bibliography
“Alternative Financial Services: A Primer”. FDIC Quarterly 2009, Volume 3: Number 1. Pages
39-47.
https://www.fdic.gov/bank/analytical/quarterly/2009_vol3_1/FDIC140_QuarterlyVol3No
1_AFS_FINAL.pdf
“FDIC Survey of Banks’ Efforts to Serve the Unbanked and Underbanked: Executive Summary
of Findings and Recommendations.” February, 2009.
https://www.fdic.gov/unbankedsurveys/unbankedstudy/FDICBankSurvey_ExecSummary
“Financial Assets and Income: Consumers with Subprime Credit.” 2015 CFED Survey.
http://scorecard.assetsandopportunity.org/latest/measure/consumers-with-subprime-
credit.
“The Impact of Differences between consumer- and creditor-purchased credit scores.” CFPB.
July 19, 2011.
http://files.consumerfinance.gov/f/2011/07/Report_20110719_CreditScores.pdf
107 Congress Public Law 204 (Sarbanes-Oxley). http://www.gpo.gov/fdsys/pkg/PLAW-
107publ204/html/PLAW-107publ204.htm. 2002.
12 CFR 202—Equal Credit Opportunity (Regulation B). Office of the Comptroller of the
Currency. U.S. Department of Treasury. July 15, 2011. http://www.occ.gov/news-
issuances/bulletins/2011/bulletin-2011-39.html.
12 U.S. Code Paragraph 2901. https://www.law.cornell.edu/uscode/text/12/2901.
15 U.S.C. 1601 et seq.; 12 C.F.R. pts. 226 and 1026
15 U.S.C. 1691 et seq., 12 C.F.R. pts. 202 and 1002 and 12 C.F.R. 701.31 (NCUA)
27
16 C.F.R. Part 255, FTC “Guides Concerning the Use of Endorsements and Testimonials in
Advertising”.https://www.ftc.gov/sites/default/files/attachments/press-releases/ftc-
publishes-final-guides-governing-endorsements-
testimonials/091005revisedendorsementguides.pdf. 2009.
2013 FDIC National Survey of Unbanked and Underbanked Households. October 2014.
https://www.fdic.gov/householdsurvey/2013report.pdf
42 U.S. Code Chapter 45, Subchapter I. https://www.law.cornell.edu/uscode/text/42/chapter-
45/subchapter-I.
American Values Survey. October 29th, 2013. http://publicreligion.org/research/2013/10/2013-
american-values-survey/#.VT2jNK1Viko
Bilton, Nick. “Social Media Bots Offer Phony Friends and Real Profits.” Nov. 19th, 2014.
http://www.nytimes.com/2014/11/20/fashion/social-media-bots-offer-phony-friends-and-
real-profit.html?_r=0.
Crosman, Penny and Andy Peters. “New Underbanked FICO Score Faces Old Banker
Skepticism”. April 2, 2015. http://www.americanbanker.com/news/bank-
technology/new-underbanked-fico-score-faces-old-banker-skepticism-1073600-1.html.
Federal Deposit Insurance Corporation. https://www.fdic.gov/about/index.html#1. 2015.
Flores-Yeffal, Nadia Y. and Li Zhang. “The Role of Social Networks in Determining Earnings:
A Comparison Analysis of Four Racial and Ethnic Groups.” Sociology Mind: Vol 2, 5,
2012. 235-246. http://www.scirp.org/journal/PaperDownload.aspx?paperID=18447
Harnett, Sam. “One Uber Driver’s Story: How He Was Trapped by Auto-Loan Program.” April
24, 2015. ww2.kqed.org/news/2015/04/24/one-uber-drivers-story-how-he-was-trapped-
by-auto-loan-program.
28
Homer, Sidney. History of Interest Rates: 2000 B.C. to the Present. New Jersey: Rutgers U.P.,
1963.
Klepper, David. “NY lawmakers examine subprime auto loans.” April 23, 2015.
www.washingtontimes.com/news/2015/apr/23/ny-lawmakers-examine-subprime-auto-
loans
Labaton, Stephen. “Agency’s ’04 Rule Let Banks Pile Up New Debt”, The New York Times.
http://www.nytimes.com/2008/10/03/business/03sec.html. October 2, 2008.
Lending Club. “How does an online credit marketplace work?”
https://www.lendingclub.com/public/how-peer-lending-works.action. 2015.
Meeks, John. “FICO Scores”. FDIC. October 28th, 2011.
http://www.wvasf.org/presentation_pdfs/John_Meeks_-
_WV_Asset_Building_Charleston_102811.pdf
National Commission for the Protection of Human Subjects of Biomedical and Behavioral
Research; “Ethical Principles and Guidelines for the Protection of Human Subjects of
Research”; http://www.hhs.gov/ohrp/humansubjects/guidance/belmont.html., April 18,
1979.
Rakesh, Chitra. “Online lender AvantCredit secures $34M in funding”. Venture Beat.
http://venturebeat.com/2013/05/09/online-lender-avantcredit-secures-34m-in-funding/.
May 9, 2013.
ZestFinance, “How We Do It”. http://www.zestfinance.com/how-we-do-it.html. 2012.